Tree-sequence-level metadata

[benjeffery]

As discussed in #491 it would be useful to have top-level metadata, using the same schema and encodings as row-level metadata. #491 is already a large PR so this will be done separately.

[hyanwong]

By the way, would we ever want tree-specific metadata @benjeffery ? There's really no place for this. We would need a separate "tree" table.

[jeromekelleher]

We'd need a clear and compelling justification for this - everything is fundamentally built around the idea that individual trees aren't important, it's the nodes and edges.

[hyanwong]

Yes, I generally agree. The recent foray into BEAST trees (thanks a lot for looking into this @jeromekelleher) was my initial thought for this, but since that didn't really pan out, I can't think of any immediate justification at the moment, other than people are used to annotating specific trees with additional information. In general I agree that we should force them to think otherwise, though

[hyanwong]

For easy reference, the discussion in #495 details some of the uses of ts-level metadata, and some discussion of how we might want it to work.

[benjeffery]

Once #534 is complete, if the user really wanted per-tree metadata it could be stored as an array in the tree-sequence metadata. As this is not a common use-case I think that should suffice.

[jeromekelleher]

Good call @benjeffery, that's a very pragmatic idea.